Unfortunately, I don't know how to do draw commutative diagrams in TeX so I'll hope you're familiar with the statement of the lemma.
We want to show of course that $$0 \rightarrow \ker(f) \rightarrow \ker(g) \rightarrow \ker(h) \rightarrow \newcommand{\coker}{\operatorname{coker}}\coker(f) \rightarrow \coker(g) \rightarrow \coker(h) \rightarrow 0$$ is exact (where $f, g, h$ are as usual as in the statement of the lemma). The challenge of the snake's lemmma is to find the map $\ker(h) \rightarrow \coker(f)$ and show exactness at this step. But I'm afraid to say I don't see how even $0 \rightarrow \ker(f) \rightarrow \ker(g) \rightarrow \ker(h)$ and $\coker(f) \rightarrow \coker(g) \rightarrow \coker(h) \rightarrow 0$ is exact. As I stated, I would like to see the proof WITHOUT using elements, so without using Mitchell's embedding theorem, working in a general abelian category.
I know it arises 'naturally' from the fact that the sequences in the hypothesis of the lemma are exact, but I'm not sure when I actually try.